|| ||           || Doxygen's  

---

Generalized convolution on 3D images, with a 1D kernel

The pattern loops over all pixels in the image and performs a generalized convolution in the neighbourhood of each pixel, processing each dimension independent of the others. The result is written to the corresponding position in the result image.

A generalized convolution is like a convolution but the multiplication and addition steps are replaced with functors. The functors are called gMul (for generalized multiplication) and gAdd (for generalized addition). Using multiplication and addition as parameters in a generalized convolution thus turns it into a "normal" convolution. However, the are other possibilities. For example, using addition and minimum results in an erosion.

Instantiations of the pattern are invoked via

• HxImageRep::generalizedConvolutionK1d
• HxImageRep::genConvSeparated
• HxImageRep::genConv3dSep

Instantiation is done via the HxImgFtorGenConv3dK1d functor, with the following template parameters:

• DstImgSigT is the signature type of the destination image
• SrcImgSigT is the signature type of the source image
• KerImgSigT is the signature type of the kernel image
• PixOpT is the type of the pixel combining functor
• RedOpT is the type of the pixel reducing functor
• KernelT is the type of the kernel functor

Example(s):

• instantiation : HxInstGenConv3dK1dMulAdd_c, usage : HxUniform

For further reference :

• Method frame : HxMfGenConv
• HxImageData member : HxImageData::generalizedConvolutionK1d
• HxImageData member : HxImageData::genConv3dSep

---

Return to patterns.

---